Exactly. FB happily gives away their ML tech like Llama because what they really care about is the data that can be used to train/tune models. The ML bits are just a commodity and not really worth much (something a new wave of ML startups have yet to realize).
The open source community can improve the tech - and they can then use it on their huge amounts of text and image data.
Legal problems? Update TOS like usual(did they already?). Some might leave, most will stay.